数仓调优实践丨SQL改写消除相关子查询

本文分享自华为云社区《【调优实践】SQL改写消除相关子查询》，作者： 门前一棵葡萄树 。

一、子查询

GaussDB(DWS)根据子查询在SQL语句中的位置把子查询分成了子查询、子链接两种形式。

• 子查询SubQuery：对应于查询解析树中的范围表RangeTblEntry，更通俗一些指的是出现在FROM语句后面的独立的SELECT语句。
• 子链接SubLink：对应于查询解析树中的表达式，更通俗一些指的是出现在where/on子句、targetlist里面的语句。

1.1 非相关子查询

子查询的执行不依赖于外层父查询的任何属性值。这样子查询具有独立性，可独自求解，形成一个子查询计划先于外层的查询求解。示例：

select t1.c1,t1.c2
from t1
where t1.c1 in (
    select c2
    from t2
    where t2.c2 IN (2,3,4)
);

1.2 相关子查询

子查询的执行依赖于外层父查询的一些属性值（如下列示例t2.c1 = t1.c1条件中的t1.c1）作为内层查询的一个AND-ed条件。这样的子查询不具备独立性，需要和外层查询按分组进行求解。

select t1.c1,t1.c2
from t1
where t1.c1 in (
    select c2
    from t2
    where t2.c1 = t1.c1 AND t2.c2 in (2,3,4)
);

二、调优实战

2.1 案例：

UPDATE t1
SET (c1,c2)=(
SELECT COALESCE(t2.c1, t1.c2),c2 FROM t2 WHERE t1.i1 = t2.i1  -- 相关标量子查询
);

其中子查询SELECT COALESCE(t2.c1, t1.c2),c2 FROM t2 WHERE t1.i1 = t2.i1 依赖于外层父查询的t1表，因此属于相关子查询。执行计划：

                                                                           QUERY PLAN
----------------------------------------------------------------------------------------------------------------------------------------------------------------
  id |                   operation                   |     A-time     | A-rows | E-rows | E-distinct |  Peak Memory   | E-memory | A-width | E-width | E-costs
 ----+-----------------------------------------------+----------------+--------+--------+------------+----------------+----------+---------+---------+---------
   1 | ->  Streaming (type: GATHER)                  | 8.998          |      0 |      1 |            | 24KB           |          |         |      17 | 9.83
   2 |    ->  Update on public.t1                    | [0.086, 0.096] |      2 |      2 |            | [308KB, 308KB] |          |         |      17 | 9.74
   3 |       ->  Seq Scan on public.t1               | [0.058, 0.074] |      2 |      2 |            | [32KB, 32KB]   | 1MB      |         |      17 | 3.73
   4 |          ->  Result  [3, SubPlan 1]           | [0.033, 0.034] |      2 |     10 |            | [16KB, 16KB]   | 1MB      |         |       6 | 1.36
   5 |             ->  Materialize                   | [4.167, 4.458] |     20 |     10 |            | [16KB, 16KB]   | 16MB     | [24,24] |       6 | 1.36
   6 |                ->  Streaming(type: BROADCAST) | [4.105, 4.406] |     10 |     10 |            | [48KB, 48KB]   | 2MB      |         |       6 | 1.33
   7 |                   ->  Seq Scan on public.t2   | [0.013, 0.013] |      5 |      5 |            | [32KB, 32KB]   | 1MB      |         |       6 | 1.02
   8 |          ->  Result  [3, SubPlan 2]           | [0.006, 0.021] |      2 |     10 |            | [16KB, 16KB]   | 1MB      |         |       6 | 1.36
   9 |             ->  Materialize                   | [0.055, 0.061] |     20 |     10 |            | [16KB, 16KB]   | 16MB     | [24,24] |       6 | 1.36
  10 |                ->  Streaming(type: BROADCAST) | [0.034, 0.040] |     10 |     10 |            | [48KB, 48KB]   | 2MB      |         |       6 | 1.33
  11 |                   ->  Seq Scan on public.t2   | [0.005, 0.009] |      5 |      5 |            | [32KB, 32KB]   | 1MB      |         |       6 | 1.02

2.2 子查询消除

改写策略就是解除子查询与父查询依赖关系，改写方案参考：

UPDATE t1
SET (c1,c2)=(t3.c1,t3.c2)
FROM (
SELECT t2.i1,COALESCE(t2.c1, t1.c2) c1,t2.c2 FROM t1,t2 WHERE t1.i1 = t2.i1
)t3
WHERE t1.i1 = t3.i1;

改写后，子查询独立，不再依赖父查询中元素。执行计划：

                                                                              QUERY PLAN
----------------------------------------------------------------------------------------------------------------------------------------------------------------------
  id |                      operation                      |     A-time     | A-rows | E-rows | E-distinct |  Peak Memory   | E-memory | A-width | E-width | E-costs
 ----+-----------------------------------------------------+----------------+--------+--------+------------+----------------+----------+---------+---------+---------
   1 | ->  Streaming (type: GATHER)                        | 13.141         |      0 |      1 |            | 24KB           |          |         |      33 | 10.56
   2 |    ->  Update on public.t1                          | [6.242, 6.362] |      2 |      2 |            | [308KB, 308KB] |          |         |      33 | 10.47
   3 |       ->  Streaming(type: RESTORE)                  | [6.186, 6.310] |      2 |      2 |            | [48KB, 48KB]   | 2MB      |         |      33 | 4.46
   4 |          ->  Nested Loop (5,11)                     | [4.082, 4.801] |      2 |      2 |            | [32KB, 32KB]   | 1MB      |         |      33 | 4.44
   5 |             ->  Streaming(type: BROADCAST)          | [3.804, 4.541] |      4 |      4 |            | [48KB, 48KB]   | 2MB      |         |      27 | 2.36
   6 |                ->  Nested Loop (7,8)                | [2.972, 4.267] |      2 |      2 |            | [32KB, 32KB]   | 1MB      |         |      27 | 2.20
   7 |                   ->  Seq Scan on public.t1         | [0.010, 0.011] |      2 |      2 |            | [16KB, 16KB]   | 1MB      |         |      14 | 1.01
   8 |                   ->  Materialize                   | [2.724, 4.055] |      6 |      4 |            | [16KB, 16KB]   | 16MB     | [28,28] |      13 | 1.17
   9 |                      ->  Streaming(type: BROADCAST) | [2.667, 4.008] |      4 |      4 |            | [48KB, 48KB]   | 2MB      |         |      13 | 1.17
  10 |                         ->  Seq Scan on public.t1   | [0.008, 0.012] |      2 |      2 |            | [16KB, 16KB]   | 1MB      |         |      13 | 1.01
  11 |             ->  Materialize                         | [0.018, 0.022] |     12 |      5 |            | [16KB, 16KB]   | 16MB     | [32,32] |      14 | 2.03
  12 |                ->  Seq Scan on public.t2            | [0.007, 0.009] |      5 |      5 |            

点击关注，第一时间了解华为云新鲜技术~